free-space-cache.c 69.7 KB
Newer Older
J
Josef Bacik 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 * Copyright (C) 2008 Red Hat.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

19
#include <linux/pagemap.h>
J
Josef Bacik 已提交
20
#include <linux/sched.h>
21
#include <linux/slab.h>
22
#include <linux/math64.h>
23
#include <linux/ratelimit.h>
J
Josef Bacik 已提交
24
#include "ctree.h"
25 26
#include "free-space-cache.h"
#include "transaction.h"
27
#include "disk-io.h"
28
#include "extent_io.h"
29
#include "inode-map.h"
30

31 32
#define BITS_PER_BITMAP		(PAGE_CACHE_SIZE * 8)
#define MAX_CACHE_BYTES_PER_GIG	(32 * 1024)
J
Josef Bacik 已提交
33

34
static int link_free_space(struct btrfs_free_space_ctl *ctl,
J
Josef Bacik 已提交
35 36
			   struct btrfs_free_space *info);

37 38 39
static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
					       struct btrfs_path *path,
					       u64 offset)
40 41 42 43 44 45 46 47 48 49
{
	struct btrfs_key key;
	struct btrfs_key location;
	struct btrfs_disk_key disk_key;
	struct btrfs_free_space_header *header;
	struct extent_buffer *leaf;
	struct inode *inode = NULL;
	int ret;

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
50
	key.offset = offset;
51 52 53 54 55 56
	key.type = 0;

	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
	if (ret < 0)
		return ERR_PTR(ret);
	if (ret > 0) {
57
		btrfs_release_path(path);
58 59 60 61 62 63 64 65
		return ERR_PTR(-ENOENT);
	}

	leaf = path->nodes[0];
	header = btrfs_item_ptr(leaf, path->slots[0],
				struct btrfs_free_space_header);
	btrfs_free_space_key(leaf, header, &disk_key);
	btrfs_disk_key_to_cpu(&location, &disk_key);
66
	btrfs_release_path(path);
67 68 69 70 71 72 73 74 75 76 77

	inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
	if (!inode)
		return ERR_PTR(-ENOENT);
	if (IS_ERR(inode))
		return inode;
	if (is_bad_inode(inode)) {
		iput(inode);
		return ERR_PTR(-ENOENT);
	}

78 79
	inode->i_mapping->flags &= ~__GFP_FS;

80 81 82 83 84 85 86 87
	return inode;
}

struct inode *lookup_free_space_inode(struct btrfs_root *root,
				      struct btrfs_block_group_cache
				      *block_group, struct btrfs_path *path)
{
	struct inode *inode = NULL;
88
	u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
89 90 91 92 93 94 95 96 97 98 99 100 101

	spin_lock(&block_group->lock);
	if (block_group->inode)
		inode = igrab(block_group->inode);
	spin_unlock(&block_group->lock);
	if (inode)
		return inode;

	inode = __lookup_free_space_inode(root, path,
					  block_group->key.objectid);
	if (IS_ERR(inode))
		return inode;

102
	spin_lock(&block_group->lock);
103
	if (!((BTRFS_I(inode)->flags & flags) == flags)) {
104
		printk(KERN_INFO "Old style space inode found, converting.\n");
105 106
		BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
			BTRFS_INODE_NODATACOW;
107 108 109
		block_group->disk_cache_state = BTRFS_DC_CLEAR;
	}

110
	if (!block_group->iref) {
111 112 113 114 115 116 117 118
		block_group->inode = igrab(inode);
		block_group->iref = 1;
	}
	spin_unlock(&block_group->lock);

	return inode;
}

119 120 121
int __create_free_space_inode(struct btrfs_root *root,
			      struct btrfs_trans_handle *trans,
			      struct btrfs_path *path, u64 ino, u64 offset)
122 123 124 125 126 127
{
	struct btrfs_key key;
	struct btrfs_disk_key disk_key;
	struct btrfs_free_space_header *header;
	struct btrfs_inode_item *inode_item;
	struct extent_buffer *leaf;
128
	u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
129 130
	int ret;

131
	ret = btrfs_insert_empty_inode(trans, root, path, ino);
132 133 134
	if (ret)
		return ret;

135 136 137 138
	/* We inline crc's for the free disk space cache */
	if (ino != BTRFS_FREE_INO_OBJECTID)
		flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;

139 140 141 142 143 144 145 146 147 148 149 150
	leaf = path->nodes[0];
	inode_item = btrfs_item_ptr(leaf, path->slots[0],
				    struct btrfs_inode_item);
	btrfs_item_key(leaf, &disk_key, path->slots[0]);
	memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
			     sizeof(*inode_item));
	btrfs_set_inode_generation(leaf, inode_item, trans->transid);
	btrfs_set_inode_size(leaf, inode_item, 0);
	btrfs_set_inode_nbytes(leaf, inode_item, 0);
	btrfs_set_inode_uid(leaf, inode_item, 0);
	btrfs_set_inode_gid(leaf, inode_item, 0);
	btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
151
	btrfs_set_inode_flags(leaf, inode_item, flags);
152 153
	btrfs_set_inode_nlink(leaf, inode_item, 1);
	btrfs_set_inode_transid(leaf, inode_item, trans->transid);
154
	btrfs_set_inode_block_group(leaf, inode_item, offset);
155
	btrfs_mark_buffer_dirty(leaf);
156
	btrfs_release_path(path);
157 158

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
159
	key.offset = offset;
160 161 162 163 164
	key.type = 0;

	ret = btrfs_insert_empty_item(trans, root, path, &key,
				      sizeof(struct btrfs_free_space_header));
	if (ret < 0) {
165
		btrfs_release_path(path);
166 167 168 169 170 171 172 173
		return ret;
	}
	leaf = path->nodes[0];
	header = btrfs_item_ptr(leaf, path->slots[0],
				struct btrfs_free_space_header);
	memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
	btrfs_set_free_space_key(leaf, header, &disk_key);
	btrfs_mark_buffer_dirty(leaf);
174
	btrfs_release_path(path);
175 176 177 178

	return 0;
}

179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194
int create_free_space_inode(struct btrfs_root *root,
			    struct btrfs_trans_handle *trans,
			    struct btrfs_block_group_cache *block_group,
			    struct btrfs_path *path)
{
	int ret;
	u64 ino;

	ret = btrfs_find_free_objectid(root, &ino);
	if (ret < 0)
		return ret;

	return __create_free_space_inode(root, trans, path, ino,
					 block_group->key.objectid);
}

195 196 197 198 199
int btrfs_truncate_free_space_cache(struct btrfs_root *root,
				    struct btrfs_trans_handle *trans,
				    struct btrfs_path *path,
				    struct inode *inode)
{
200
	struct btrfs_block_rsv *rsv;
201
	u64 needed_bytes;
202 203 204
	loff_t oldsize;
	int ret = 0;

205
	rsv = trans->block_rsv;
206 207 208 209 210 211 212 213 214 215 216 217 218
	trans->block_rsv = &root->fs_info->global_block_rsv;

	/* 1 for slack space, 1 for updating the inode */
	needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
		btrfs_calc_trans_metadata_size(root, 1);

	spin_lock(&trans->block_rsv->lock);
	if (trans->block_rsv->reserved < needed_bytes) {
		spin_unlock(&trans->block_rsv->lock);
		trans->block_rsv = rsv;
		return -ENOSPC;
	}
	spin_unlock(&trans->block_rsv->lock);
219 220 221 222 223 224 225 226 227 228 229

	oldsize = i_size_read(inode);
	btrfs_i_size_write(inode, 0);
	truncate_pagecache(inode, oldsize, 0);

	/*
	 * We don't need an orphan item because truncating the free space cache
	 * will never be split across transactions.
	 */
	ret = btrfs_truncate_inode_items(trans, root, inode,
					 0, BTRFS_EXTENT_DATA_KEY);
230

231
	if (ret) {
232
		trans->block_rsv = rsv;
233 234 235 236
		WARN_ON(1);
		return ret;
	}

237
	ret = btrfs_update_inode(trans, root, inode);
238 239
	trans->block_rsv = rsv;

240
	return ret;
241 242
}

243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261
static int readahead_cache(struct inode *inode)
{
	struct file_ra_state *ra;
	unsigned long last_index;

	ra = kzalloc(sizeof(*ra), GFP_NOFS);
	if (!ra)
		return -ENOMEM;

	file_ra_state_init(ra, inode->i_mapping);
	last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;

	page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);

	kfree(ra);

	return 0;
}

262 263 264 265 266 267 268 269
struct io_ctl {
	void *cur, *orig;
	struct page *page;
	struct page **pages;
	struct btrfs_root *root;
	unsigned long size;
	int index;
	int num_pages;
270
	unsigned check_crcs:1;
271 272 273 274 275 276 277 278 279 280 281 282 283
};

static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
		       struct btrfs_root *root)
{
	memset(io_ctl, 0, sizeof(struct io_ctl));
	io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
		PAGE_CACHE_SHIFT;
	io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
				GFP_NOFS);
	if (!io_ctl->pages)
		return -ENOMEM;
	io_ctl->root = root;
284 285
	if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
		io_ctl->check_crcs = 1;
286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353
	return 0;
}

static void io_ctl_free(struct io_ctl *io_ctl)
{
	kfree(io_ctl->pages);
}

static void io_ctl_unmap_page(struct io_ctl *io_ctl)
{
	if (io_ctl->cur) {
		kunmap(io_ctl->page);
		io_ctl->cur = NULL;
		io_ctl->orig = NULL;
	}
}

static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
{
	WARN_ON(io_ctl->cur);
	BUG_ON(io_ctl->index >= io_ctl->num_pages);
	io_ctl->page = io_ctl->pages[io_ctl->index++];
	io_ctl->cur = kmap(io_ctl->page);
	io_ctl->orig = io_ctl->cur;
	io_ctl->size = PAGE_CACHE_SIZE;
	if (clear)
		memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
}

static void io_ctl_drop_pages(struct io_ctl *io_ctl)
{
	int i;

	io_ctl_unmap_page(io_ctl);

	for (i = 0; i < io_ctl->num_pages; i++) {
		ClearPageChecked(io_ctl->pages[i]);
		unlock_page(io_ctl->pages[i]);
		page_cache_release(io_ctl->pages[i]);
	}
}

static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
				int uptodate)
{
	struct page *page;
	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
	int i;

	for (i = 0; i < io_ctl->num_pages; i++) {
		page = find_or_create_page(inode->i_mapping, i, mask);
		if (!page) {
			io_ctl_drop_pages(io_ctl);
			return -ENOMEM;
		}
		io_ctl->pages[i] = page;
		if (uptodate && !PageUptodate(page)) {
			btrfs_readpage(NULL, page);
			lock_page(page);
			if (!PageUptodate(page)) {
				printk(KERN_ERR "btrfs: error reading free "
				       "space cache\n");
				io_ctl_drop_pages(io_ctl);
				return -EIO;
			}
		}
	}

354 355 356 357 358
	for (i = 0; i < io_ctl->num_pages; i++) {
		clear_page_dirty_for_io(io_ctl->pages[i]);
		set_page_extent_mapped(io_ctl->pages[i]);
	}

359 360 361 362 363 364 365 366 367 368
	return 0;
}

static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
{
	u64 *val;

	io_ctl_map_page(io_ctl, 1);

	/*
369 370
	 * Skip the csum areas.  If we don't check crcs then we just have a
	 * 64bit chunk at the front of the first page.
371
	 */
372 373 374 375 376 377 378
	if (io_ctl->check_crcs) {
		io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
		io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
	} else {
		io_ctl->cur += sizeof(u64);
		io_ctl->size -= sizeof(u64) * 2;
	}
379 380 381 382 383 384 385 386 387 388

	val = io_ctl->cur;
	*val = cpu_to_le64(generation);
	io_ctl->cur += sizeof(u64);
}

static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
{
	u64 *gen;

389 390 391 392 393 394 395 396 397 398 399 400
	/*
	 * Skip the crc area.  If we don't check crcs then we just have a 64bit
	 * chunk at the front of the first page.
	 */
	if (io_ctl->check_crcs) {
		io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
		io_ctl->size -= sizeof(u64) +
			(sizeof(u32) * io_ctl->num_pages);
	} else {
		io_ctl->cur += sizeof(u64);
		io_ctl->size -= sizeof(u64) * 2;
	}
401 402 403 404 405 406 407 408 409 410

	gen = io_ctl->cur;
	if (le64_to_cpu(*gen) != generation) {
		printk_ratelimited(KERN_ERR "btrfs: space cache generation "
				   "(%Lu) does not match inode (%Lu)\n", *gen,
				   generation);
		io_ctl_unmap_page(io_ctl);
		return -EIO;
	}
	io_ctl->cur += sizeof(u64);
411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467
	return 0;
}

static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
{
	u32 *tmp;
	u32 crc = ~(u32)0;
	unsigned offset = 0;

	if (!io_ctl->check_crcs) {
		io_ctl_unmap_page(io_ctl);
		return;
	}

	if (index == 0)
		offset = sizeof(u32) * io_ctl->num_pages;;

	crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
			      PAGE_CACHE_SIZE - offset);
	btrfs_csum_final(crc, (char *)&crc);
	io_ctl_unmap_page(io_ctl);
	tmp = kmap(io_ctl->pages[0]);
	tmp += index;
	*tmp = crc;
	kunmap(io_ctl->pages[0]);
}

static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
{
	u32 *tmp, val;
	u32 crc = ~(u32)0;
	unsigned offset = 0;

	if (!io_ctl->check_crcs) {
		io_ctl_map_page(io_ctl, 0);
		return 0;
	}

	if (index == 0)
		offset = sizeof(u32) * io_ctl->num_pages;

	tmp = kmap(io_ctl->pages[0]);
	tmp += index;
	val = *tmp;
	kunmap(io_ctl->pages[0]);

	io_ctl_map_page(io_ctl, 0);
	crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
			      PAGE_CACHE_SIZE - offset);
	btrfs_csum_final(crc, (char *)&crc);
	if (val != crc) {
		printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
				   "space cache\n");
		io_ctl_unmap_page(io_ctl);
		return -EIO;
	}

468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489
	return 0;
}

static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
			    void *bitmap)
{
	struct btrfs_free_space_entry *entry;

	if (!io_ctl->cur)
		return -ENOSPC;

	entry = io_ctl->cur;
	entry->offset = cpu_to_le64(offset);
	entry->bytes = cpu_to_le64(bytes);
	entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
		BTRFS_FREE_SPACE_EXTENT;
	io_ctl->cur += sizeof(struct btrfs_free_space_entry);
	io_ctl->size -= sizeof(struct btrfs_free_space_entry);

	if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
		return 0;

490
	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510

	/* No more pages to map */
	if (io_ctl->index >= io_ctl->num_pages)
		return 0;

	/* map the next page */
	io_ctl_map_page(io_ctl, 1);
	return 0;
}

static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
{
	if (!io_ctl->cur)
		return -ENOSPC;

	/*
	 * If we aren't at the start of the current page, unmap this one and
	 * map the next one if there is any left.
	 */
	if (io_ctl->cur != io_ctl->orig) {
511
		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
512 513 514 515 516 517
		if (io_ctl->index >= io_ctl->num_pages)
			return -ENOSPC;
		io_ctl_map_page(io_ctl, 0);
	}

	memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
518
	io_ctl_set_crc(io_ctl, io_ctl->index - 1);
519 520 521 522 523 524 525
	if (io_ctl->index < io_ctl->num_pages)
		io_ctl_map_page(io_ctl, 0);
	return 0;
}

static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
{
526 527 528 529 530 531 532 533
	/*
	 * If we're not on the boundary we know we've modified the page and we
	 * need to crc the page.
	 */
	if (io_ctl->cur != io_ctl->orig)
		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
	else
		io_ctl_unmap_page(io_ctl);
534 535 536

	while (io_ctl->index < io_ctl->num_pages) {
		io_ctl_map_page(io_ctl, 1);
537
		io_ctl_set_crc(io_ctl, io_ctl->index - 1);
538 539 540
	}
}

541 542
static int io_ctl_read_entry(struct io_ctl *io_ctl,
			    struct btrfs_free_space *entry, u8 *type)
543 544
{
	struct btrfs_free_space_entry *e;
545 546 547 548 549 550 551
	int ret;

	if (!io_ctl->cur) {
		ret = io_ctl_check_crc(io_ctl, io_ctl->index);
		if (ret)
			return ret;
	}
552 553 554 555

	e = io_ctl->cur;
	entry->offset = le64_to_cpu(e->offset);
	entry->bytes = le64_to_cpu(e->bytes);
556
	*type = e->type;
557 558 559 560
	io_ctl->cur += sizeof(struct btrfs_free_space_entry);
	io_ctl->size -= sizeof(struct btrfs_free_space_entry);

	if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
561
		return 0;
562 563 564

	io_ctl_unmap_page(io_ctl);

565
	return 0;
566 567
}

568 569
static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
			      struct btrfs_free_space *entry)
570
{
571 572 573 574 575 576
	int ret;

	ret = io_ctl_check_crc(io_ctl, io_ctl->index);
	if (ret)
		return ret;

577 578
	memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
	io_ctl_unmap_page(io_ctl);
579 580

	return 0;
581 582
}

583 584 585
int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
			    struct btrfs_free_space_ctl *ctl,
			    struct btrfs_path *path, u64 offset)
586 587 588
{
	struct btrfs_free_space_header *header;
	struct extent_buffer *leaf;
589
	struct io_ctl io_ctl;
590
	struct btrfs_key key;
591
	struct btrfs_free_space *e, *n;
592 593 594 595
	struct list_head bitmaps;
	u64 num_entries;
	u64 num_bitmaps;
	u64 generation;
596
	u8 type;
597
	int ret = 0;
598 599 600 601

	INIT_LIST_HEAD(&bitmaps);

	/* Nothing in the space cache, goodbye */
602
	if (!i_size_read(inode))
603
		return 0;
604 605

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
606
	key.offset = offset;
607 608 609
	key.type = 0;

	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
610
	if (ret < 0)
611
		return 0;
612
	else if (ret > 0) {
613
		btrfs_release_path(path);
614
		return 0;
615 616
	}

617 618
	ret = -1;

619 620 621 622 623 624
	leaf = path->nodes[0];
	header = btrfs_item_ptr(leaf, path->slots[0],
				struct btrfs_free_space_header);
	num_entries = btrfs_free_space_entries(leaf, header);
	num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
	generation = btrfs_free_space_generation(leaf, header);
625
	btrfs_release_path(path);
626 627 628

	if (BTRFS_I(inode)->generation != generation) {
		printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
629
		       " not match free space cache generation (%llu)\n",
630
		       (unsigned long long)BTRFS_I(inode)->generation,
631
		       (unsigned long long)generation);
632
		return 0;
633 634 635
	}

	if (!num_entries)
636
		return 0;
637

638
	io_ctl_init(&io_ctl, inode, root);
639
	ret = readahead_cache(inode);
640
	if (ret)
641 642
		goto out;

643 644 645
	ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
	if (ret)
		goto out;
646

647 648 649 650
	ret = io_ctl_check_crc(&io_ctl, 0);
	if (ret)
		goto free_cache;

651 652 653
	ret = io_ctl_check_generation(&io_ctl, generation);
	if (ret)
		goto free_cache;
654

655 656 657 658
	while (num_entries) {
		e = kmem_cache_zalloc(btrfs_free_space_cachep,
				      GFP_NOFS);
		if (!e)
659 660
			goto free_cache;

661 662 663 664 665 666
		ret = io_ctl_read_entry(&io_ctl, e, &type);
		if (ret) {
			kmem_cache_free(btrfs_free_space_cachep, e);
			goto free_cache;
		}

667 668 669
		if (!e->bytes) {
			kmem_cache_free(btrfs_free_space_cachep, e);
			goto free_cache;
670
		}
671 672 673 674 675 676 677 678 679

		if (type == BTRFS_FREE_SPACE_EXTENT) {
			spin_lock(&ctl->tree_lock);
			ret = link_free_space(ctl, e);
			spin_unlock(&ctl->tree_lock);
			if (ret) {
				printk(KERN_ERR "Duplicate entries in "
				       "free space cache, dumping\n");
				kmem_cache_free(btrfs_free_space_cachep, e);
680 681
				goto free_cache;
			}
682 683 684 685 686 687 688
		} else {
			BUG_ON(!num_bitmaps);
			num_bitmaps--;
			e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
			if (!e->bitmap) {
				kmem_cache_free(
					btrfs_free_space_cachep, e);
689 690
				goto free_cache;
			}
691 692 693 694 695 696 697 698
			spin_lock(&ctl->tree_lock);
			ret = link_free_space(ctl, e);
			ctl->total_bitmaps++;
			ctl->op->recalc_thresholds(ctl);
			spin_unlock(&ctl->tree_lock);
			if (ret) {
				printk(KERN_ERR "Duplicate entries in "
				       "free space cache, dumping\n");
699
				kmem_cache_free(btrfs_free_space_cachep, e);
700 701
				goto free_cache;
			}
702
			list_add_tail(&e->list, &bitmaps);
703 704
		}

705 706
		num_entries--;
	}
707

708 709
	io_ctl_unmap_page(&io_ctl);

710 711 712 713 714
	/*
	 * We add the bitmaps at the end of the entries in order that
	 * the bitmap entries are added to the cache.
	 */
	list_for_each_entry_safe(e, n, &bitmaps, list) {
715
		list_del_init(&e->list);
716 717 718
		ret = io_ctl_read_bitmap(&io_ctl, e);
		if (ret)
			goto free_cache;
719 720
	}

721
	io_ctl_drop_pages(&io_ctl);
722 723
	ret = 1;
out:
724
	io_ctl_free(&io_ctl);
725 726
	return ret;
free_cache:
727
	io_ctl_drop_pages(&io_ctl);
728
	__btrfs_remove_free_space_cache(ctl);
729 730 731
	goto out;
}

732 733
int load_free_space_cache(struct btrfs_fs_info *fs_info,
			  struct btrfs_block_group_cache *block_group)
J
Josef Bacik 已提交
734
{
735
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
736 737 738
	struct btrfs_root *root = fs_info->tree_root;
	struct inode *inode;
	struct btrfs_path *path;
739
	int ret = 0;
740 741 742 743 744 745 746
	bool matched;
	u64 used = btrfs_block_group_used(&block_group->item);

	/*
	 * If we're unmounting then just return, since this does a search on the
	 * normal root and not the commit root and we could deadlock.
	 */
747
	if (btrfs_fs_closing(fs_info))
748 749 750 751 752 753
		return 0;

	/*
	 * If this block group has been marked to be cleared for one reason or
	 * another then we can't trust the on disk cache, so just return.
	 */
754
	spin_lock(&block_group->lock);
755 756 757 758
	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
		spin_unlock(&block_group->lock);
		return 0;
	}
759
	spin_unlock(&block_group->lock);
760 761 762 763 764 765 766 767 768 769 770

	path = btrfs_alloc_path();
	if (!path)
		return 0;

	inode = lookup_free_space_inode(root, block_group, path);
	if (IS_ERR(inode)) {
		btrfs_free_path(path);
		return 0;
	}

771 772 773 774 775 776 777 778
	/* We may have converted the inode and made the cache invalid. */
	spin_lock(&block_group->lock);
	if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
		spin_unlock(&block_group->lock);
		goto out;
	}
	spin_unlock(&block_group->lock);

779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801
	ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
				      path, block_group->key.objectid);
	btrfs_free_path(path);
	if (ret <= 0)
		goto out;

	spin_lock(&ctl->tree_lock);
	matched = (ctl->free_space == (block_group->key.offset - used -
				       block_group->bytes_super));
	spin_unlock(&ctl->tree_lock);

	if (!matched) {
		__btrfs_remove_free_space_cache(ctl);
		printk(KERN_ERR "block group %llu has an wrong amount of free "
		       "space\n", block_group->key.objectid);
		ret = -1;
	}
out:
	if (ret < 0) {
		/* This cache is bogus, make sure it gets cleared */
		spin_lock(&block_group->lock);
		block_group->disk_cache_state = BTRFS_DC_CLEAR;
		spin_unlock(&block_group->lock);
802
		ret = 0;
803 804 805 806 807 808 809

		printk(KERN_ERR "btrfs: failed to load free space cache "
		       "for block group %llu\n", block_group->key.objectid);
	}

	iput(inode);
	return ret;
810 811
}

812 813 814 815 816 817 818 819 820 821 822 823 824
/**
 * __btrfs_write_out_cache - write out cached info to an inode
 * @root - the root the inode belongs to
 * @ctl - the free space cache we are going to write out
 * @block_group - the block_group for this cache if it belongs to a block_group
 * @trans - the trans handle
 * @path - the path to use
 * @offset - the offset for the key we'll insert
 *
 * This function writes out a free space cache struct to disk for quick recovery
 * on mount.  This will return 0 if it was successfull in writing the cache out,
 * and -1 if it was not.
 */
825 826 827 828 829
int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
			    struct btrfs_free_space_ctl *ctl,
			    struct btrfs_block_group_cache *block_group,
			    struct btrfs_trans_handle *trans,
			    struct btrfs_path *path, u64 offset)
J
Josef Bacik 已提交
830 831 832 833 834 835
{
	struct btrfs_free_space_header *header;
	struct extent_buffer *leaf;
	struct rb_node *node;
	struct list_head *pos, *n;
	struct extent_state *cached_state = NULL;
836 837
	struct btrfs_free_cluster *cluster = NULL;
	struct extent_io_tree *unpin = NULL;
838
	struct io_ctl io_ctl;
J
Josef Bacik 已提交
839 840
	struct list_head bitmap_list;
	struct btrfs_key key;
841
	u64 start, end, len;
J
Josef Bacik 已提交
842 843
	int entries = 0;
	int bitmaps = 0;
844 845
	int ret;
	int err = -1;
J
Josef Bacik 已提交
846 847 848

	INIT_LIST_HEAD(&bitmap_list);

849 850
	if (!i_size_read(inode))
		return -1;
851

852
	io_ctl_init(&io_ctl, inode, root);
853

854
	/* Get the cluster for this block_group if it exists */
855
	if (block_group && !list_empty(&block_group->cluster_list))
856 857 858 859 860 861 862 863 864 865
		cluster = list_entry(block_group->cluster_list.next,
				     struct btrfs_free_cluster,
				     block_group_list);

	/*
	 * We shouldn't have switched the pinned extents yet so this is the
	 * right one
	 */
	unpin = root->fs_info->pinned_extents;

866 867
	/* Lock all pages first so we can lock the extent safely. */
	io_ctl_prepare_pages(&io_ctl, inode, 0);
J
Josef Bacik 已提交
868 869 870 871

	lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
			 0, &cached_state, GFP_NOFS);

872 873 874 875
	/*
	 * When searching for pinned extents, we need to start at our start
	 * offset.
	 */
876 877
	if (block_group)
		start = block_group->key.objectid;
878

879 880 881 882 883 884
	node = rb_first(&ctl->free_space_offset);
	if (!node && cluster) {
		node = rb_first(&cluster->root);
		cluster = NULL;
	}

885 886 887 888 889 890 891
	/* Make sure we can fit our crcs into the first page */
	if (io_ctl.check_crcs &&
	    (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) {
		WARN_ON(1);
		goto out_nospc;
	}

892
	io_ctl_set_generation(&io_ctl, trans->transid);
893

894 895 896
	/* Write out the extent entries */
	while (node) {
		struct btrfs_free_space *e;
J
Josef Bacik 已提交
897

898 899
		e = rb_entry(node, struct btrfs_free_space, offset_index);
		entries++;
J
Josef Bacik 已提交
900

901 902 903 904
		ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
				       e->bitmap);
		if (ret)
			goto out_nospc;
905

906 907 908
		if (e->bitmap) {
			list_add_tail(&e->list, &bitmap_list);
			bitmaps++;
909
		}
910 911 912 913
		node = rb_next(node);
		if (!node && cluster) {
			node = rb_first(&cluster->root);
			cluster = NULL;
914
		}
915
	}
916

917 918 919 920 921 922 923 924 925 926 927
	/*
	 * We want to add any pinned extents to our free space cache
	 * so we don't leak the space
	 */
	while (block_group && (start < block_group->key.objectid +
			       block_group->key.offset)) {
		ret = find_first_extent_bit(unpin, start, &start, &end,
					    EXTENT_DIRTY);
		if (ret) {
			ret = 0;
			break;
J
Josef Bacik 已提交
928 929
		}

930 931 932 933
		/* This pinned extent is out of our range */
		if (start >= block_group->key.objectid +
		    block_group->key.offset)
			break;
934

935 936 937
		len = block_group->key.objectid +
			block_group->key.offset - start;
		len = min(len, end + 1 - start);
J
Josef Bacik 已提交
938

939 940 941 942
		entries++;
		ret = io_ctl_add_entry(&io_ctl, start, len, NULL);
		if (ret)
			goto out_nospc;
J
Josef Bacik 已提交
943

944 945
		start = end + 1;
	}
J
Josef Bacik 已提交
946 947 948 949 950 951

	/* Write out the bitmaps */
	list_for_each_safe(pos, n, &bitmap_list) {
		struct btrfs_free_space *entry =
			list_entry(pos, struct btrfs_free_space, list);

952 953 954
		ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
		if (ret)
			goto out_nospc;
J
Josef Bacik 已提交
955
		list_del_init(&entry->list);
956 957
	}

J
Josef Bacik 已提交
958
	/* Zero out the rest of the pages just to make sure */
959
	io_ctl_zero_remaining_pages(&io_ctl);
J
Josef Bacik 已提交
960

961 962 963
	ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
				0, i_size_read(inode), &cached_state);
	io_ctl_drop_pages(&io_ctl);
J
Josef Bacik 已提交
964 965 966
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
			     i_size_read(inode) - 1, &cached_state, GFP_NOFS);

967
	if (ret)
968
		goto out;
969 970


971 972 973
	ret = filemap_write_and_wait(inode->i_mapping);
	if (ret)
		goto out;
J
Josef Bacik 已提交
974 975

	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
976
	key.offset = offset;
J
Josef Bacik 已提交
977 978
	key.type = 0;

979
	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
J
Josef Bacik 已提交
980
	if (ret < 0) {
981
		clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
982 983
				 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
				 GFP_NOFS);
984
		goto out;
J
Josef Bacik 已提交
985 986 987 988 989 990 991 992
	}
	leaf = path->nodes[0];
	if (ret > 0) {
		struct btrfs_key found_key;
		BUG_ON(!path->slots[0]);
		path->slots[0]--;
		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
		if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
993
		    found_key.offset != offset) {
994 995
			clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
					 inode->i_size - 1,
996 997
					 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
					 NULL, GFP_NOFS);
998
			btrfs_release_path(path);
999
			goto out;
J
Josef Bacik 已提交
1000 1001
		}
	}
1002 1003

	BTRFS_I(inode)->generation = trans->transid;
J
Josef Bacik 已提交
1004 1005 1006 1007 1008 1009
	header = btrfs_item_ptr(leaf, path->slots[0],
				struct btrfs_free_space_header);
	btrfs_set_free_space_entries(leaf, header, entries);
	btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
	btrfs_set_free_space_generation(leaf, header, trans->transid);
	btrfs_mark_buffer_dirty(leaf);
1010
	btrfs_release_path(path);
J
Josef Bacik 已提交
1011

1012
	err = 0;
1013
out:
1014
	io_ctl_free(&io_ctl);
1015
	if (err) {
1016
		invalidate_inode_pages2(inode->i_mapping);
J
Josef Bacik 已提交
1017 1018 1019
		BTRFS_I(inode)->generation = 0;
	}
	btrfs_update_inode(trans, root, inode);
1020
	return err;
1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031

out_nospc:
	list_for_each_safe(pos, n, &bitmap_list) {
		struct btrfs_free_space *entry =
			list_entry(pos, struct btrfs_free_space, list);
		list_del_init(&entry->list);
	}
	io_ctl_drop_pages(&io_ctl);
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
			     i_size_read(inode) - 1, &cached_state, GFP_NOFS);
	goto out;
1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057
}

int btrfs_write_out_cache(struct btrfs_root *root,
			  struct btrfs_trans_handle *trans,
			  struct btrfs_block_group_cache *block_group,
			  struct btrfs_path *path)
{
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
	struct inode *inode;
	int ret = 0;

	root = root->fs_info->tree_root;

	spin_lock(&block_group->lock);
	if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
		spin_unlock(&block_group->lock);
		return 0;
	}
	spin_unlock(&block_group->lock);

	inode = lookup_free_space_inode(root, block_group, path);
	if (IS_ERR(inode))
		return 0;

	ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
				      path, block_group->key.objectid);
1058
	if (ret) {
1059 1060 1061
		spin_lock(&block_group->lock);
		block_group->disk_cache_state = BTRFS_DC_ERROR;
		spin_unlock(&block_group->lock);
1062
		ret = 0;
1063
#ifdef DEBUG
1064 1065
		printk(KERN_ERR "btrfs: failed to write free space cace "
		       "for block group %llu\n", block_group->key.objectid);
1066
#endif
1067 1068
	}

J
Josef Bacik 已提交
1069 1070 1071 1072
	iput(inode);
	return ret;
}

1073
static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1074
					  u64 offset)
J
Josef Bacik 已提交
1075
{
1076 1077
	BUG_ON(offset < bitmap_start);
	offset -= bitmap_start;
1078
	return (unsigned long)(div_u64(offset, unit));
1079
}
J
Josef Bacik 已提交
1080

1081
static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1082
{
1083
	return (unsigned long)(div_u64(bytes, unit));
1084
}
J
Josef Bacik 已提交
1085

1086
static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1087 1088 1089 1090
				   u64 offset)
{
	u64 bitmap_start;
	u64 bytes_per_bitmap;
J
Josef Bacik 已提交
1091

1092 1093
	bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
	bitmap_start = offset - ctl->start;
1094 1095
	bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
	bitmap_start *= bytes_per_bitmap;
1096
	bitmap_start += ctl->start;
J
Josef Bacik 已提交
1097

1098
	return bitmap_start;
J
Josef Bacik 已提交
1099 1100
}

1101 1102
static int tree_insert_offset(struct rb_root *root, u64 offset,
			      struct rb_node *node, int bitmap)
J
Josef Bacik 已提交
1103 1104 1105 1106 1107 1108 1109
{
	struct rb_node **p = &root->rb_node;
	struct rb_node *parent = NULL;
	struct btrfs_free_space *info;

	while (*p) {
		parent = *p;
1110
		info = rb_entry(parent, struct btrfs_free_space, offset_index);
J
Josef Bacik 已提交
1111

1112
		if (offset < info->offset) {
J
Josef Bacik 已提交
1113
			p = &(*p)->rb_left;
1114
		} else if (offset > info->offset) {
J
Josef Bacik 已提交
1115
			p = &(*p)->rb_right;
1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130
		} else {
			/*
			 * we could have a bitmap entry and an extent entry
			 * share the same offset.  If this is the case, we want
			 * the extent entry to always be found first if we do a
			 * linear search through the tree, since we want to have
			 * the quickest allocation time, and allocating from an
			 * extent is faster than allocating from a bitmap.  So
			 * if we're inserting a bitmap and we find an entry at
			 * this offset, we want to go right, or after this entry
			 * logically.  If we are inserting an extent and we've
			 * found a bitmap, we want to go left, or before
			 * logically.
			 */
			if (bitmap) {
1131 1132 1133 1134
				if (info->bitmap) {
					WARN_ON_ONCE(1);
					return -EEXIST;
				}
1135 1136
				p = &(*p)->rb_right;
			} else {
1137 1138 1139 1140
				if (!info->bitmap) {
					WARN_ON_ONCE(1);
					return -EEXIST;
				}
1141 1142 1143
				p = &(*p)->rb_left;
			}
		}
J
Josef Bacik 已提交
1144 1145 1146 1147 1148 1149 1150 1151 1152
	}

	rb_link_node(node, parent, p);
	rb_insert_color(node, root);

	return 0;
}

/*
J
Josef Bacik 已提交
1153 1154
 * searches the tree for the given offset.
 *
1155 1156 1157
 * fuzzy - If this is set, then we are trying to make an allocation, and we just
 * want a section that has at least bytes size and comes at or after the given
 * offset.
J
Josef Bacik 已提交
1158
 */
1159
static struct btrfs_free_space *
1160
tree_search_offset(struct btrfs_free_space_ctl *ctl,
1161
		   u64 offset, int bitmap_only, int fuzzy)
J
Josef Bacik 已提交
1162
{
1163
	struct rb_node *n = ctl->free_space_offset.rb_node;
1164 1165 1166 1167 1168 1169 1170 1171
	struct btrfs_free_space *entry, *prev = NULL;

	/* find entry that is closest to the 'offset' */
	while (1) {
		if (!n) {
			entry = NULL;
			break;
		}
J
Josef Bacik 已提交
1172 1173

		entry = rb_entry(n, struct btrfs_free_space, offset_index);
1174
		prev = entry;
J
Josef Bacik 已提交
1175

1176
		if (offset < entry->offset)
J
Josef Bacik 已提交
1177
			n = n->rb_left;
1178
		else if (offset > entry->offset)
J
Josef Bacik 已提交
1179
			n = n->rb_right;
1180
		else
J
Josef Bacik 已提交
1181 1182 1183
			break;
	}

1184 1185 1186 1187 1188
	if (bitmap_only) {
		if (!entry)
			return NULL;
		if (entry->bitmap)
			return entry;
J
Josef Bacik 已提交
1189

1190 1191 1192 1193 1194 1195 1196 1197 1198 1199
		/*
		 * bitmap entry and extent entry may share same offset,
		 * in that case, bitmap entry comes after extent entry.
		 */
		n = rb_next(n);
		if (!n)
			return NULL;
		entry = rb_entry(n, struct btrfs_free_space, offset_index);
		if (entry->offset != offset)
			return NULL;
J
Josef Bacik 已提交
1200

1201 1202 1203 1204
		WARN_ON(!entry->bitmap);
		return entry;
	} else if (entry) {
		if (entry->bitmap) {
J
Josef Bacik 已提交
1205
			/*
1206 1207
			 * if previous extent entry covers the offset,
			 * we should return it instead of the bitmap entry
J
Josef Bacik 已提交
1208
			 */
1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220
			n = &entry->offset_index;
			while (1) {
				n = rb_prev(n);
				if (!n)
					break;
				prev = rb_entry(n, struct btrfs_free_space,
						offset_index);
				if (!prev->bitmap) {
					if (prev->offset + prev->bytes > offset)
						entry = prev;
					break;
				}
J
Josef Bacik 已提交
1221
			}
1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236
		}
		return entry;
	}

	if (!prev)
		return NULL;

	/* find last entry before the 'offset' */
	entry = prev;
	if (entry->offset > offset) {
		n = rb_prev(&entry->offset_index);
		if (n) {
			entry = rb_entry(n, struct btrfs_free_space,
					offset_index);
			BUG_ON(entry->offset > offset);
J
Josef Bacik 已提交
1237
		} else {
1238 1239 1240 1241
			if (fuzzy)
				return entry;
			else
				return NULL;
J
Josef Bacik 已提交
1242 1243 1244
		}
	}

1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258
	if (entry->bitmap) {
		n = &entry->offset_index;
		while (1) {
			n = rb_prev(n);
			if (!n)
				break;
			prev = rb_entry(n, struct btrfs_free_space,
					offset_index);
			if (!prev->bitmap) {
				if (prev->offset + prev->bytes > offset)
					return prev;
				break;
			}
		}
1259
		if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1260 1261 1262 1263 1264 1265 1266 1267 1268 1269
			return entry;
	} else if (entry->offset + entry->bytes > offset)
		return entry;

	if (!fuzzy)
		return NULL;

	while (1) {
		if (entry->bitmap) {
			if (entry->offset + BITS_PER_BITMAP *
1270
			    ctl->unit > offset)
1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282
				break;
		} else {
			if (entry->offset + entry->bytes > offset)
				break;
		}

		n = rb_next(&entry->offset_index);
		if (!n)
			return NULL;
		entry = rb_entry(n, struct btrfs_free_space, offset_index);
	}
	return entry;
J
Josef Bacik 已提交
1283 1284
}

1285
static inline void
1286
__unlink_free_space(struct btrfs_free_space_ctl *ctl,
1287
		    struct btrfs_free_space *info)
J
Josef Bacik 已提交
1288
{
1289 1290
	rb_erase(&info->offset_index, &ctl->free_space_offset);
	ctl->free_extents--;
1291 1292
}

1293
static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1294 1295
			      struct btrfs_free_space *info)
{
1296 1297
	__unlink_free_space(ctl, info);
	ctl->free_space -= info->bytes;
J
Josef Bacik 已提交
1298 1299
}

1300
static int link_free_space(struct btrfs_free_space_ctl *ctl,
J
Josef Bacik 已提交
1301 1302 1303 1304
			   struct btrfs_free_space *info)
{
	int ret = 0;

1305
	BUG_ON(!info->bitmap && !info->bytes);
1306
	ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1307
				 &info->offset_index, (info->bitmap != NULL));
J
Josef Bacik 已提交
1308 1309 1310
	if (ret)
		return ret;

1311 1312
	ctl->free_space += info->bytes;
	ctl->free_extents++;
1313 1314 1315
	return ret;
}

1316
static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1317
{
1318
	struct btrfs_block_group_cache *block_group = ctl->private;
1319 1320 1321
	u64 max_bytes;
	u64 bitmap_bytes;
	u64 extent_bytes;
1322
	u64 size = block_group->key.offset;
1323 1324 1325 1326
	u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize;
	int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);

	BUG_ON(ctl->total_bitmaps > max_bitmaps);
1327 1328 1329 1330 1331 1332

	/*
	 * The goal is to keep the total amount of memory used per 1gb of space
	 * at or below 32k, so we need to adjust how much memory we allow to be
	 * used by extent based free space tracking
	 */
1333 1334 1335 1336 1337
	if (size < 1024 * 1024 * 1024)
		max_bytes = MAX_CACHE_BYTES_PER_GIG;
	else
		max_bytes = MAX_CACHE_BYTES_PER_GIG *
			div64_u64(size, 1024 * 1024 * 1024);
1338

1339 1340 1341 1342 1343
	/*
	 * we want to account for 1 more bitmap than what we have so we can make
	 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
	 * we add more bitmaps.
	 */
1344
	bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1345

1346
	if (bitmap_bytes >= max_bytes) {
1347
		ctl->extents_thresh = 0;
1348 1349
		return;
	}
1350

1351 1352 1353 1354 1355 1356
	/*
	 * we want the extent entry threshold to always be at most 1/2 the maxw
	 * bytes we can have, or whatever is less than that.
	 */
	extent_bytes = max_bytes - bitmap_bytes;
	extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1357

1358
	ctl->extents_thresh =
1359
		div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1360 1361
}

1362 1363 1364
static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
				       struct btrfs_free_space *info,
				       u64 offset, u64 bytes)
1365
{
L
Li Zefan 已提交
1366
	unsigned long start, count;
1367

1368 1369
	start = offset_to_bit(info->offset, ctl->unit, offset);
	count = bytes_to_bits(bytes, ctl->unit);
L
Li Zefan 已提交
1370
	BUG_ON(start + count > BITS_PER_BITMAP);
1371

L
Li Zefan 已提交
1372
	bitmap_clear(info->bitmap, start, count);
1373 1374

	info->bytes -= bytes;
1375 1376 1377 1378 1379 1380 1381
}

static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
			      struct btrfs_free_space *info, u64 offset,
			      u64 bytes)
{
	__bitmap_clear_bits(ctl, info, offset, bytes);
1382
	ctl->free_space -= bytes;
1383 1384
}

1385
static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
J
Josef Bacik 已提交
1386 1387
			    struct btrfs_free_space *info, u64 offset,
			    u64 bytes)
1388
{
L
Li Zefan 已提交
1389
	unsigned long start, count;
1390

1391 1392
	start = offset_to_bit(info->offset, ctl->unit, offset);
	count = bytes_to_bits(bytes, ctl->unit);
L
Li Zefan 已提交
1393
	BUG_ON(start + count > BITS_PER_BITMAP);
1394

L
Li Zefan 已提交
1395
	bitmap_set(info->bitmap, start, count);
1396 1397

	info->bytes += bytes;
1398
	ctl->free_space += bytes;
1399 1400
}

1401
static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1402 1403 1404 1405 1406 1407 1408
			 struct btrfs_free_space *bitmap_info, u64 *offset,
			 u64 *bytes)
{
	unsigned long found_bits = 0;
	unsigned long bits, i;
	unsigned long next_zero;

1409
	i = offset_to_bit(bitmap_info->offset, ctl->unit,
1410
			  max_t(u64, *offset, bitmap_info->offset));
1411
	bits = bytes_to_bits(*bytes, ctl->unit);
1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425

	for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i);
	     i < BITS_PER_BITMAP;
	     i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) {
		next_zero = find_next_zero_bit(bitmap_info->bitmap,
					       BITS_PER_BITMAP, i);
		if ((next_zero - i) >= bits) {
			found_bits = next_zero - i;
			break;
		}
		i = next_zero;
	}

	if (found_bits) {
1426 1427
		*offset = (u64)(i * ctl->unit) + bitmap_info->offset;
		*bytes = (u64)(found_bits) * ctl->unit;
1428 1429 1430 1431 1432 1433
		return 0;
	}

	return -1;
}

1434 1435
static struct btrfs_free_space *
find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes)
1436 1437 1438 1439 1440
{
	struct btrfs_free_space *entry;
	struct rb_node *node;
	int ret;

1441
	if (!ctl->free_space_offset.rb_node)
1442 1443
		return NULL;

1444
	entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1445 1446 1447 1448 1449 1450 1451 1452 1453
	if (!entry)
		return NULL;

	for (node = &entry->offset_index; node; node = rb_next(node)) {
		entry = rb_entry(node, struct btrfs_free_space, offset_index);
		if (entry->bytes < *bytes)
			continue;

		if (entry->bitmap) {
1454
			ret = search_bitmap(ctl, entry, offset, bytes);
1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467
			if (!ret)
				return entry;
			continue;
		}

		*offset = entry->offset;
		*bytes = entry->bytes;
		return entry;
	}

	return NULL;
}

1468
static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1469 1470
			   struct btrfs_free_space *info, u64 offset)
{
1471
	info->offset = offset_to_bitmap(ctl, offset);
J
Josef Bacik 已提交
1472
	info->bytes = 0;
1473
	INIT_LIST_HEAD(&info->list);
1474 1475
	link_free_space(ctl, info);
	ctl->total_bitmaps++;
1476

1477
	ctl->op->recalc_thresholds(ctl);
1478 1479
}

1480
static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1481 1482
			struct btrfs_free_space *bitmap_info)
{
1483
	unlink_free_space(ctl, bitmap_info);
1484
	kfree(bitmap_info->bitmap);
1485
	kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1486 1487
	ctl->total_bitmaps--;
	ctl->op->recalc_thresholds(ctl);
1488 1489
}

1490
static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1491 1492 1493 1494
			      struct btrfs_free_space *bitmap_info,
			      u64 *offset, u64 *bytes)
{
	u64 end;
1495 1496
	u64 search_start, search_bytes;
	int ret;
1497 1498

again:
1499
	end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1500

1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512
	/*
	 * XXX - this can go away after a few releases.
	 *
	 * since the only user of btrfs_remove_free_space is the tree logging
	 * stuff, and the only way to test that is under crash conditions, we
	 * want to have this debug stuff here just in case somethings not
	 * working.  Search the bitmap for the space we are trying to use to
	 * make sure its actually there.  If its not there then we need to stop
	 * because something has gone wrong.
	 */
	search_start = *offset;
	search_bytes = *bytes;
1513
	search_bytes = min(search_bytes, end - search_start + 1);
1514
	ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1515 1516
	BUG_ON(ret < 0 || search_start != *offset);

1517
	if (*offset > bitmap_info->offset && *offset + *bytes > end) {
1518
		bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1);
1519 1520 1521
		*bytes -= end - *offset + 1;
		*offset = end + 1;
	} else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) {
1522
		bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes);
1523 1524 1525 1526
		*bytes = 0;
	}

	if (*bytes) {
1527
		struct rb_node *next = rb_next(&bitmap_info->offset_index);
1528
		if (!bitmap_info->bytes)
1529
			free_bitmap(ctl, bitmap_info);
1530

1531 1532 1533 1534 1535
		/*
		 * no entry after this bitmap, but we still have bytes to
		 * remove, so something has gone wrong.
		 */
		if (!next)
1536 1537
			return -EINVAL;

1538 1539 1540 1541 1542 1543 1544
		bitmap_info = rb_entry(next, struct btrfs_free_space,
				       offset_index);

		/*
		 * if the next entry isn't a bitmap we need to return to let the
		 * extent stuff do its work.
		 */
1545 1546 1547
		if (!bitmap_info->bitmap)
			return -EAGAIN;

1548 1549 1550 1551 1552 1553 1554 1555
		/*
		 * Ok the next item is a bitmap, but it may not actually hold
		 * the information for the rest of this free space stuff, so
		 * look for it, and if we don't find it return so we can try
		 * everything over again.
		 */
		search_start = *offset;
		search_bytes = *bytes;
1556
		ret = search_bitmap(ctl, bitmap_info, &search_start,
1557 1558 1559 1560
				    &search_bytes);
		if (ret < 0 || search_start != *offset)
			return -EAGAIN;

1561
		goto again;
1562
	} else if (!bitmap_info->bytes)
1563
		free_bitmap(ctl, bitmap_info);
1564 1565 1566 1567

	return 0;
}

J
Josef Bacik 已提交
1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584
static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
			       struct btrfs_free_space *info, u64 offset,
			       u64 bytes)
{
	u64 bytes_to_set = 0;
	u64 end;

	end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);

	bytes_to_set = min(end - offset, bytes);

	bitmap_set_bits(ctl, info, offset, bytes_to_set);

	return bytes_to_set;

}

1585 1586
static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
		      struct btrfs_free_space *info)
1587
{
1588
	struct btrfs_block_group_cache *block_group = ctl->private;
1589 1590 1591 1592 1593

	/*
	 * If we are below the extents threshold then we can add this as an
	 * extent, and don't have to deal with the bitmap
	 */
1594
	if (ctl->free_extents < ctl->extents_thresh) {
1595 1596 1597 1598 1599 1600 1601 1602
		/*
		 * If this block group has some small extents we don't want to
		 * use up all of our free slots in the cache with them, we want
		 * to reserve them to larger extents, however if we have plent
		 * of cache left then go ahead an dadd them, no sense in adding
		 * the overhead of a bitmap if we don't have to.
		 */
		if (info->bytes <= block_group->sectorsize * 4) {
1603 1604
			if (ctl->free_extents * 2 <= ctl->extents_thresh)
				return false;
1605
		} else {
1606
			return false;
1607 1608
		}
	}
1609 1610 1611 1612 1613 1614 1615

	/*
	 * some block groups are so tiny they can't be enveloped by a bitmap, so
	 * don't even bother to create a bitmap for this
	 */
	if (BITS_PER_BITMAP * block_group->sectorsize >
	    block_group->key.offset)
1616 1617 1618 1619 1620
		return false;

	return true;
}

J
Josef Bacik 已提交
1621 1622 1623 1624 1625
static struct btrfs_free_space_op free_space_op = {
	.recalc_thresholds	= recalculate_thresholds,
	.use_bitmap		= use_bitmap,
};

1626 1627 1628 1629
static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
			      struct btrfs_free_space *info)
{
	struct btrfs_free_space *bitmap_info;
J
Josef Bacik 已提交
1630
	struct btrfs_block_group_cache *block_group = NULL;
1631
	int added = 0;
J
Josef Bacik 已提交
1632
	u64 bytes, offset, bytes_added;
1633
	int ret;
1634 1635 1636 1637

	bytes = info->bytes;
	offset = info->offset;

1638 1639 1640
	if (!ctl->op->use_bitmap(ctl, info))
		return 0;

J
Josef Bacik 已提交
1641 1642
	if (ctl->op == &free_space_op)
		block_group = ctl->private;
1643
again:
J
Josef Bacik 已提交
1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660
	/*
	 * Since we link bitmaps right into the cluster we need to see if we
	 * have a cluster here, and if so and it has our bitmap we need to add
	 * the free space to that bitmap.
	 */
	if (block_group && !list_empty(&block_group->cluster_list)) {
		struct btrfs_free_cluster *cluster;
		struct rb_node *node;
		struct btrfs_free_space *entry;

		cluster = list_entry(block_group->cluster_list.next,
				     struct btrfs_free_cluster,
				     block_group_list);
		spin_lock(&cluster->lock);
		node = rb_first(&cluster->root);
		if (!node) {
			spin_unlock(&cluster->lock);
1661
			goto no_cluster_bitmap;
J
Josef Bacik 已提交
1662 1663 1664 1665 1666
		}

		entry = rb_entry(node, struct btrfs_free_space, offset_index);
		if (!entry->bitmap) {
			spin_unlock(&cluster->lock);
1667
			goto no_cluster_bitmap;
J
Josef Bacik 已提交
1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681
		}

		if (entry->offset == offset_to_bitmap(ctl, offset)) {
			bytes_added = add_bytes_to_bitmap(ctl, entry,
							  offset, bytes);
			bytes -= bytes_added;
			offset += bytes_added;
		}
		spin_unlock(&cluster->lock);
		if (!bytes) {
			ret = 1;
			goto out;
		}
	}
1682 1683

no_cluster_bitmap:
1684
	bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1685 1686 1687 1688 1689 1690
					 1, 0);
	if (!bitmap_info) {
		BUG_ON(added);
		goto new_bitmap;
	}

J
Josef Bacik 已提交
1691 1692 1693 1694
	bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
	bytes -= bytes_added;
	offset += bytes_added;
	added = 0;
1695 1696 1697 1698 1699 1700 1701 1702 1703

	if (!bytes) {
		ret = 1;
		goto out;
	} else
		goto again;

new_bitmap:
	if (info && info->bitmap) {
1704
		add_new_bitmap(ctl, info, offset);
1705 1706 1707 1708
		added = 1;
		info = NULL;
		goto again;
	} else {
1709
		spin_unlock(&ctl->tree_lock);
1710 1711 1712

		/* no pre-allocated info, allocate a new one */
		if (!info) {
1713 1714
			info = kmem_cache_zalloc(btrfs_free_space_cachep,
						 GFP_NOFS);
1715
			if (!info) {
1716
				spin_lock(&ctl->tree_lock);
1717 1718 1719 1720 1721 1722 1723
				ret = -ENOMEM;
				goto out;
			}
		}

		/* allocate the bitmap */
		info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1724
		spin_lock(&ctl->tree_lock);
1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735
		if (!info->bitmap) {
			ret = -ENOMEM;
			goto out;
		}
		goto again;
	}

out:
	if (info) {
		if (info->bitmap)
			kfree(info->bitmap);
1736
		kmem_cache_free(btrfs_free_space_cachep, info);
1737
	}
J
Josef Bacik 已提交
1738 1739 1740 1741

	return ret;
}

1742
static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1743
			  struct btrfs_free_space *info, bool update_stat)
J
Josef Bacik 已提交
1744
{
1745 1746 1747 1748 1749
	struct btrfs_free_space *left_info;
	struct btrfs_free_space *right_info;
	bool merged = false;
	u64 offset = info->offset;
	u64 bytes = info->bytes;
1750

J
Josef Bacik 已提交
1751 1752 1753 1754 1755
	/*
	 * first we want to see if there is free space adjacent to the range we
	 * are adding, if there is remove that struct and add a new one to
	 * cover the entire range
	 */
1756
	right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1757 1758 1759 1760
	if (right_info && rb_prev(&right_info->offset_index))
		left_info = rb_entry(rb_prev(&right_info->offset_index),
				     struct btrfs_free_space, offset_index);
	else
1761
		left_info = tree_search_offset(ctl, offset - 1, 0, 0);
J
Josef Bacik 已提交
1762

1763
	if (right_info && !right_info->bitmap) {
1764
		if (update_stat)
1765
			unlink_free_space(ctl, right_info);
1766
		else
1767
			__unlink_free_space(ctl, right_info);
1768
		info->bytes += right_info->bytes;
1769
		kmem_cache_free(btrfs_free_space_cachep, right_info);
1770
		merged = true;
J
Josef Bacik 已提交
1771 1772
	}

1773 1774
	if (left_info && !left_info->bitmap &&
	    left_info->offset + left_info->bytes == offset) {
1775
		if (update_stat)
1776
			unlink_free_space(ctl, left_info);
1777
		else
1778
			__unlink_free_space(ctl, left_info);
1779 1780
		info->offset = left_info->offset;
		info->bytes += left_info->bytes;
1781
		kmem_cache_free(btrfs_free_space_cachep, left_info);
1782
		merged = true;
J
Josef Bacik 已提交
1783 1784
	}

1785 1786 1787
	return merged;
}

1788 1789
int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
			   u64 offset, u64 bytes)
1790 1791 1792 1793
{
	struct btrfs_free_space *info;
	int ret = 0;

1794
	info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1795 1796 1797 1798 1799 1800
	if (!info)
		return -ENOMEM;

	info->offset = offset;
	info->bytes = bytes;

1801
	spin_lock(&ctl->tree_lock);
1802

1803
	if (try_merge_free_space(ctl, info, true))
1804 1805 1806 1807 1808 1809 1810
		goto link;

	/*
	 * There was no extent directly to the left or right of this new
	 * extent then we know we're going to have to allocate a new extent, so
	 * before we do that see if we need to drop this into a bitmap
	 */
1811
	ret = insert_into_bitmap(ctl, info);
1812 1813 1814 1815 1816 1817 1818
	if (ret < 0) {
		goto out;
	} else if (ret) {
		ret = 0;
		goto out;
	}
link:
1819
	ret = link_free_space(ctl, info);
J
Josef Bacik 已提交
1820
	if (ret)
1821
		kmem_cache_free(btrfs_free_space_cachep, info);
1822
out:
1823
	spin_unlock(&ctl->tree_lock);
1824

J
Josef Bacik 已提交
1825
	if (ret) {
1826
		printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
S
Stoyan Gaydarov 已提交
1827
		BUG_ON(ret == -EEXIST);
J
Josef Bacik 已提交
1828 1829 1830 1831 1832
	}

	return ret;
}

1833 1834
int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
			    u64 offset, u64 bytes)
J
Josef Bacik 已提交
1835
{
1836
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
J
Josef Bacik 已提交
1837
	struct btrfs_free_space *info;
1838
	struct btrfs_free_space *next_info = NULL;
J
Josef Bacik 已提交
1839 1840
	int ret = 0;

1841
	spin_lock(&ctl->tree_lock);
1842

1843
again:
1844
	info = tree_search_offset(ctl, offset, 0, 0);
1845
	if (!info) {
1846 1847 1848 1849
		/*
		 * oops didn't find an extent that matched the space we wanted
		 * to remove, look for a bitmap instead
		 */
1850
		info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1851 1852
					  1, 0);
		if (!info) {
1853 1854 1855 1856 1857 1858 1859
			/* the tree logging code might be calling us before we
			 * have fully loaded the free space rbtree for this
			 * block group.  So it is possible the entry won't
			 * be in the rbtree yet at all.  The caching code
			 * will make sure not to put it in the rbtree if
			 * the logging code has pinned it.
			 */
1860 1861
			goto out_lock;
		}
1862 1863 1864 1865 1866 1867 1868 1869 1870
	}

	if (info->bytes < bytes && rb_next(&info->offset_index)) {
		u64 end;
		next_info = rb_entry(rb_next(&info->offset_index),
					     struct btrfs_free_space,
					     offset_index);

		if (next_info->bitmap)
1871 1872
			end = next_info->offset +
			      BITS_PER_BITMAP * ctl->unit - 1;
1873 1874 1875 1876 1877 1878 1879 1880 1881 1882
		else
			end = next_info->offset + next_info->bytes;

		if (next_info->bytes < bytes ||
		    next_info->offset > offset || offset > end) {
			printk(KERN_CRIT "Found free space at %llu, size %llu,"
			      " trying to use %llu\n",
			      (unsigned long long)info->offset,
			      (unsigned long long)info->bytes,
			      (unsigned long long)bytes);
J
Josef Bacik 已提交
1883 1884
			WARN_ON(1);
			ret = -EINVAL;
1885
			goto out_lock;
J
Josef Bacik 已提交
1886 1887
		}

1888 1889 1890 1891
		info = next_info;
	}

	if (info->bytes == bytes) {
1892
		unlink_free_space(ctl, info);
1893 1894
		if (info->bitmap) {
			kfree(info->bitmap);
1895
			ctl->total_bitmaps--;
J
Josef Bacik 已提交
1896
		}
1897
		kmem_cache_free(btrfs_free_space_cachep, info);
1898
		ret = 0;
1899 1900
		goto out_lock;
	}
J
Josef Bacik 已提交
1901

1902
	if (!info->bitmap && info->offset == offset) {
1903
		unlink_free_space(ctl, info);
J
Josef Bacik 已提交
1904 1905
		info->offset += bytes;
		info->bytes -= bytes;
1906 1907
		ret = link_free_space(ctl, info);
		WARN_ON(ret);
1908 1909
		goto out_lock;
	}
J
Josef Bacik 已提交
1910

1911 1912
	if (!info->bitmap && info->offset <= offset &&
	    info->offset + info->bytes >= offset + bytes) {
1913 1914 1915 1916 1917 1918 1919 1920
		u64 old_start = info->offset;
		/*
		 * we're freeing space in the middle of the info,
		 * this can happen during tree log replay
		 *
		 * first unlink the old info and then
		 * insert it again after the hole we're creating
		 */
1921
		unlink_free_space(ctl, info);
1922 1923 1924 1925 1926
		if (offset + bytes < info->offset + info->bytes) {
			u64 old_end = info->offset + info->bytes;

			info->offset = offset + bytes;
			info->bytes = old_end - info->offset;
1927
			ret = link_free_space(ctl, info);
1928 1929 1930
			WARN_ON(ret);
			if (ret)
				goto out_lock;
1931 1932 1933 1934
		} else {
			/* the hole we're creating ends at the end
			 * of the info struct, just free the info
			 */
1935
			kmem_cache_free(btrfs_free_space_cachep, info);
1936
		}
1937
		spin_unlock(&ctl->tree_lock);
1938 1939 1940

		/* step two, insert a new info struct to cover
		 * anything before the hole
1941
		 */
1942 1943
		ret = btrfs_add_free_space(block_group, old_start,
					   offset - old_start);
1944 1945
		WARN_ON(ret);
		goto out;
J
Josef Bacik 已提交
1946
	}
1947

1948
	ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1949 1950 1951 1952
	if (ret == -EAGAIN)
		goto again;
	BUG_ON(ret);
out_lock:
1953
	spin_unlock(&ctl->tree_lock);
J
Josef Bacik 已提交
1954
out:
1955 1956 1957
	return ret;
}

J
Josef Bacik 已提交
1958 1959 1960
void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
			   u64 bytes)
{
1961
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
J
Josef Bacik 已提交
1962 1963 1964 1965
	struct btrfs_free_space *info;
	struct rb_node *n;
	int count = 0;

1966
	for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
J
Josef Bacik 已提交
1967 1968 1969
		info = rb_entry(n, struct btrfs_free_space, offset_index);
		if (info->bytes >= bytes)
			count++;
1970
		printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1971
		       (unsigned long long)info->offset,
1972 1973
		       (unsigned long long)info->bytes,
		       (info->bitmap) ? "yes" : "no");
J
Josef Bacik 已提交
1974
	}
1975 1976
	printk(KERN_INFO "block group has cluster?: %s\n",
	       list_empty(&block_group->cluster_list) ? "no" : "yes");
J
Josef Bacik 已提交
1977 1978 1979 1980
	printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
	       "\n", count);
}

1981
void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
J
Josef Bacik 已提交
1982
{
1983
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
J
Josef Bacik 已提交
1984

1985 1986 1987 1988 1989
	spin_lock_init(&ctl->tree_lock);
	ctl->unit = block_group->sectorsize;
	ctl->start = block_group->key.objectid;
	ctl->private = block_group;
	ctl->op = &free_space_op;
J
Josef Bacik 已提交
1990

1991 1992 1993 1994 1995 1996 1997
	/*
	 * we only want to have 32k of ram per block group for keeping
	 * track of free space, and if we pass 1/2 of that we want to
	 * start converting things over to using bitmaps
	 */
	ctl->extents_thresh = ((1024 * 32) / 2) /
				sizeof(struct btrfs_free_space);
J
Josef Bacik 已提交
1998 1999
}

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
/*
 * for a given cluster, put all of its extents back into the free
 * space cache.  If the block group passed doesn't match the block group
 * pointed to by the cluster, someone else raced in and freed the
 * cluster already.  In that case, we just return without changing anything
 */
static int
__btrfs_return_cluster_to_free_space(
			     struct btrfs_block_group_cache *block_group,
			     struct btrfs_free_cluster *cluster)
{
2011
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2012 2013 2014 2015 2016 2017 2018
	struct btrfs_free_space *entry;
	struct rb_node *node;

	spin_lock(&cluster->lock);
	if (cluster->block_group != block_group)
		goto out;

2019
	cluster->block_group = NULL;
2020
	cluster->window_start = 0;
2021 2022
	list_del_init(&cluster->block_group_list);

2023
	node = rb_first(&cluster->root);
2024
	while (node) {
2025 2026
		bool bitmap;

2027 2028 2029
		entry = rb_entry(node, struct btrfs_free_space, offset_index);
		node = rb_next(&entry->offset_index);
		rb_erase(&entry->offset_index, &cluster->root);
2030 2031 2032

		bitmap = (entry->bitmap != NULL);
		if (!bitmap)
2033 2034
			try_merge_free_space(ctl, entry, false);
		tree_insert_offset(&ctl->free_space_offset,
2035
				   entry->offset, &entry->offset_index, bitmap);
2036
	}
2037
	cluster->root = RB_ROOT;
2038

2039 2040
out:
	spin_unlock(&cluster->lock);
2041
	btrfs_put_block_group(block_group);
2042 2043 2044
	return 0;
}

2045
void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl)
J
Josef Bacik 已提交
2046 2047 2048
{
	struct btrfs_free_space *info;
	struct rb_node *node;
2049 2050 2051

	while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
		info = rb_entry(node, struct btrfs_free_space, offset_index);
2052 2053 2054 2055 2056 2057
		if (!info->bitmap) {
			unlink_free_space(ctl, info);
			kmem_cache_free(btrfs_free_space_cachep, info);
		} else {
			free_bitmap(ctl, info);
		}
2058 2059 2060 2061 2062 2063
		if (need_resched()) {
			spin_unlock(&ctl->tree_lock);
			cond_resched();
			spin_lock(&ctl->tree_lock);
		}
	}
2064 2065 2066 2067 2068 2069
}

void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
{
	spin_lock(&ctl->tree_lock);
	__btrfs_remove_free_space_cache_locked(ctl);
2070 2071 2072 2073 2074 2075
	spin_unlock(&ctl->tree_lock);
}

void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
{
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2076
	struct btrfs_free_cluster *cluster;
2077
	struct list_head *head;
J
Josef Bacik 已提交
2078

2079
	spin_lock(&ctl->tree_lock);
2080 2081 2082 2083
	while ((head = block_group->cluster_list.next) !=
	       &block_group->cluster_list) {
		cluster = list_entry(head, struct btrfs_free_cluster,
				     block_group_list);
2084 2085 2086

		WARN_ON(cluster->block_group != block_group);
		__btrfs_return_cluster_to_free_space(block_group, cluster);
2087
		if (need_resched()) {
2088
			spin_unlock(&ctl->tree_lock);
2089
			cond_resched();
2090
			spin_lock(&ctl->tree_lock);
2091
		}
2092
	}
2093
	__btrfs_remove_free_space_cache_locked(ctl);
2094
	spin_unlock(&ctl->tree_lock);
2095

J
Josef Bacik 已提交
2096 2097
}

2098 2099
u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
			       u64 offset, u64 bytes, u64 empty_size)
J
Josef Bacik 已提交
2100
{
2101
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2102
	struct btrfs_free_space *entry = NULL;
2103
	u64 bytes_search = bytes + empty_size;
2104
	u64 ret = 0;
J
Josef Bacik 已提交
2105

2106 2107
	spin_lock(&ctl->tree_lock);
	entry = find_free_space(ctl, &offset, &bytes_search);
2108
	if (!entry)
2109 2110 2111 2112
		goto out;

	ret = offset;
	if (entry->bitmap) {
2113
		bitmap_clear_bits(ctl, entry, offset, bytes);
2114
		if (!entry->bytes)
2115
			free_bitmap(ctl, entry);
2116
	} else {
2117
		unlink_free_space(ctl, entry);
2118 2119 2120
		entry->offset += bytes;
		entry->bytes -= bytes;
		if (!entry->bytes)
2121
			kmem_cache_free(btrfs_free_space_cachep, entry);
2122
		else
2123
			link_free_space(ctl, entry);
2124
	}
J
Josef Bacik 已提交
2125

2126
out:
2127
	spin_unlock(&ctl->tree_lock);
J
Josef Bacik 已提交
2128

J
Josef Bacik 已提交
2129 2130
	return ret;
}
2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143

/*
 * given a cluster, put all of its extents back into the free space
 * cache.  If a block group is passed, this function will only free
 * a cluster that belongs to the passed block group.
 *
 * Otherwise, it'll get a reference on the block group pointed to by the
 * cluster and remove the cluster from it.
 */
int btrfs_return_cluster_to_free_space(
			       struct btrfs_block_group_cache *block_group,
			       struct btrfs_free_cluster *cluster)
{
2144
	struct btrfs_free_space_ctl *ctl;
2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162
	int ret;

	/* first, get a safe pointer to the block group */
	spin_lock(&cluster->lock);
	if (!block_group) {
		block_group = cluster->block_group;
		if (!block_group) {
			spin_unlock(&cluster->lock);
			return 0;
		}
	} else if (cluster->block_group != block_group) {
		/* someone else has already freed it don't redo their work */
		spin_unlock(&cluster->lock);
		return 0;
	}
	atomic_inc(&block_group->count);
	spin_unlock(&cluster->lock);

2163 2164
	ctl = block_group->free_space_ctl;

2165
	/* now return any extents the cluster had on it */
2166
	spin_lock(&ctl->tree_lock);
2167
	ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2168
	spin_unlock(&ctl->tree_lock);
2169 2170 2171 2172 2173 2174

	/* finally drop our ref */
	btrfs_put_block_group(block_group);
	return ret;
}

2175 2176
static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
				   struct btrfs_free_cluster *cluster,
2177
				   struct btrfs_free_space *entry,
2178 2179
				   u64 bytes, u64 min_start)
{
2180
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2181 2182 2183 2184 2185 2186 2187 2188
	int err;
	u64 search_start = cluster->window_start;
	u64 search_bytes = bytes;
	u64 ret = 0;

	search_start = min_start;
	search_bytes = bytes;

2189
	err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2190
	if (err)
2191
		return 0;
2192 2193

	ret = search_start;
2194
	__bitmap_clear_bits(ctl, entry, ret, bytes);
2195 2196 2197 2198

	return ret;
}

2199 2200 2201 2202 2203 2204 2205 2206 2207
/*
 * given a cluster, try to allocate 'bytes' from it, returns 0
 * if it couldn't find anything suitably large, or a logical disk offset
 * if things worked out
 */
u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
			     struct btrfs_free_cluster *cluster, u64 bytes,
			     u64 min_start)
{
2208
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225
	struct btrfs_free_space *entry = NULL;
	struct rb_node *node;
	u64 ret = 0;

	spin_lock(&cluster->lock);
	if (bytes > cluster->max_size)
		goto out;

	if (cluster->block_group != block_group)
		goto out;

	node = rb_first(&cluster->root);
	if (!node)
		goto out;

	entry = rb_entry(node, struct btrfs_free_space, offset_index);
	while(1) {
2226 2227
		if (entry->bytes < bytes ||
		    (!entry->bitmap && entry->offset < min_start)) {
2228 2229 2230 2231 2232 2233 2234 2235
			node = rb_next(&entry->offset_index);
			if (!node)
				break;
			entry = rb_entry(node, struct btrfs_free_space,
					 offset_index);
			continue;
		}

2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253
		if (entry->bitmap) {
			ret = btrfs_alloc_from_bitmap(block_group,
						      cluster, entry, bytes,
						      min_start);
			if (ret == 0) {
				node = rb_next(&entry->offset_index);
				if (!node)
					break;
				entry = rb_entry(node, struct btrfs_free_space,
						 offset_index);
				continue;
			}
		} else {
			ret = entry->offset;

			entry->offset += bytes;
			entry->bytes -= bytes;
		}
2254

2255
		if (entry->bytes == 0)
2256 2257 2258 2259 2260
			rb_erase(&entry->offset_index, &cluster->root);
		break;
	}
out:
	spin_unlock(&cluster->lock);
2261

2262 2263 2264
	if (!ret)
		return 0;

2265
	spin_lock(&ctl->tree_lock);
2266

2267
	ctl->free_space -= bytes;
2268
	if (entry->bytes == 0) {
2269
		ctl->free_extents--;
2270 2271
		if (entry->bitmap) {
			kfree(entry->bitmap);
2272 2273
			ctl->total_bitmaps--;
			ctl->op->recalc_thresholds(ctl);
2274
		}
2275
		kmem_cache_free(btrfs_free_space_cachep, entry);
2276 2277
	}

2278
	spin_unlock(&ctl->tree_lock);
2279

2280 2281 2282
	return ret;
}

2283 2284 2285
static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
				struct btrfs_free_space *entry,
				struct btrfs_free_cluster *cluster,
2286 2287
				u64 offset, u64 bytes,
				u64 cont1_bytes, u64 min_bytes)
2288
{
2289
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2290 2291
	unsigned long next_zero;
	unsigned long i;
2292 2293
	unsigned long want_bits;
	unsigned long min_bits;
2294 2295 2296
	unsigned long found_bits;
	unsigned long start = 0;
	unsigned long total_found = 0;
2297
	int ret;
2298 2299 2300

	i = offset_to_bit(entry->offset, block_group->sectorsize,
			  max_t(u64, offset, entry->offset));
2301 2302
	want_bits = bytes_to_bits(bytes, block_group->sectorsize);
	min_bits = bytes_to_bits(min_bytes, block_group->sectorsize);
2303 2304 2305 2306 2307 2308 2309 2310

again:
	found_bits = 0;
	for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i);
	     i < BITS_PER_BITMAP;
	     i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) {
		next_zero = find_next_zero_bit(entry->bitmap,
					       BITS_PER_BITMAP, i);
2311
		if (next_zero - i >= min_bits) {
2312 2313 2314 2315 2316 2317 2318
			found_bits = next_zero - i;
			break;
		}
		i = next_zero;
	}

	if (!found_bits)
2319
		return -ENOSPC;
2320

2321
	if (!total_found) {
2322
		start = i;
2323
		cluster->max_size = 0;
2324 2325 2326 2327 2328 2329 2330
	}

	total_found += found_bits;

	if (cluster->max_size < found_bits * block_group->sectorsize)
		cluster->max_size = found_bits * block_group->sectorsize;

2331 2332
	if (total_found < want_bits || cluster->max_size < cont1_bytes) {
		i = next_zero + 1;
2333 2334 2335 2336 2337
		goto again;
	}

	cluster->window_start = start * block_group->sectorsize +
		entry->offset;
2338
	rb_erase(&entry->offset_index, &ctl->free_space_offset);
2339 2340 2341
	ret = tree_insert_offset(&cluster->root, entry->offset,
				 &entry->offset_index, 1);
	BUG_ON(ret);
2342 2343 2344 2345

	return 0;
}

2346 2347
/*
 * This searches the block group for just extents to fill the cluster with.
2348 2349
 * Try to find a cluster with at least bytes total bytes, at least one
 * extent of cont1_bytes, and other clusters of at least min_bytes.
2350
 */
2351 2352 2353 2354
static noinline int
setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
			struct btrfs_free_cluster *cluster,
			struct list_head *bitmaps, u64 offset, u64 bytes,
2355
			u64 cont1_bytes, u64 min_bytes)
2356
{
2357
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2358 2359 2360 2361 2362 2363 2364 2365
	struct btrfs_free_space *first = NULL;
	struct btrfs_free_space *entry = NULL;
	struct btrfs_free_space *last;
	struct rb_node *node;
	u64 window_start;
	u64 window_free;
	u64 max_extent;

2366
	entry = tree_search_offset(ctl, offset, 0, 1);
2367 2368 2369 2370 2371 2372 2373
	if (!entry)
		return -ENOSPC;

	/*
	 * We don't want bitmaps, so just move along until we find a normal
	 * extent entry.
	 */
2374 2375
	while (entry->bitmap || entry->bytes < min_bytes) {
		if (entry->bitmap && list_empty(&entry->list))
2376
			list_add_tail(&entry->list, bitmaps);
2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388
		node = rb_next(&entry->offset_index);
		if (!node)
			return -ENOSPC;
		entry = rb_entry(node, struct btrfs_free_space, offset_index);
	}

	window_start = entry->offset;
	window_free = entry->bytes;
	max_extent = entry->bytes;
	first = entry;
	last = entry;

2389 2390
	for (node = rb_next(&entry->offset_index); node;
	     node = rb_next(&entry->offset_index)) {
2391 2392
		entry = rb_entry(node, struct btrfs_free_space, offset_index);

2393 2394 2395
		if (entry->bitmap) {
			if (list_empty(&entry->list))
				list_add_tail(&entry->list, bitmaps);
2396
			continue;
2397 2398
		}

2399 2400 2401 2402 2403 2404
		if (entry->bytes < min_bytes)
			continue;

		last = entry;
		window_free += entry->bytes;
		if (entry->bytes > max_extent)
2405 2406 2407
			max_extent = entry->bytes;
	}

2408 2409 2410
	if (window_free < bytes || max_extent < cont1_bytes)
		return -ENOSPC;

2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423
	cluster->window_start = first->offset;

	node = &first->offset_index;

	/*
	 * now we've found our entries, pull them out of the free space
	 * cache and put them into the cluster rbtree
	 */
	do {
		int ret;

		entry = rb_entry(node, struct btrfs_free_space, offset_index);
		node = rb_next(&entry->offset_index);
2424
		if (entry->bitmap || entry->bytes < min_bytes)
2425 2426
			continue;

2427
		rb_erase(&entry->offset_index, &ctl->free_space_offset);
2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441
		ret = tree_insert_offset(&cluster->root, entry->offset,
					 &entry->offset_index, 0);
		BUG_ON(ret);
	} while (node && entry != last);

	cluster->max_size = max_extent;

	return 0;
}

/*
 * This specifically looks for bitmaps that may work in the cluster, we assume
 * that we have already failed to find extents that will work.
 */
2442 2443 2444 2445
static noinline int
setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
		     struct btrfs_free_cluster *cluster,
		     struct list_head *bitmaps, u64 offset, u64 bytes,
2446
		     u64 cont1_bytes, u64 min_bytes)
2447
{
2448
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2449 2450
	struct btrfs_free_space *entry;
	int ret = -ENOSPC;
2451
	u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2452

2453
	if (ctl->total_bitmaps == 0)
2454 2455
		return -ENOSPC;

2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466
	/*
	 * The bitmap that covers offset won't be in the list unless offset
	 * is just its start offset.
	 */
	entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
	if (entry->offset != bitmap_offset) {
		entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
		if (entry && list_empty(&entry->list))
			list_add(&entry->list, bitmaps);
	}

2467 2468 2469 2470
	list_for_each_entry(entry, bitmaps, list) {
		if (entry->bytes < min_bytes)
			continue;
		ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2471
					   bytes, cont1_bytes, min_bytes);
2472 2473 2474 2475 2476
		if (!ret)
			return 0;
	}

	/*
2477 2478
	 * The bitmaps list has all the bitmaps that record free space
	 * starting after offset, so no more search is required.
2479
	 */
2480
	return -ENOSPC;
2481 2482
}

2483 2484
/*
 * here we try to find a cluster of blocks in a block group.  The goal
2485
 * is to find at least bytes+empty_size.
2486 2487 2488 2489 2490 2491
 * We might not find them all in one contiguous area.
 *
 * returns zero and sets up cluster if things worked out, otherwise
 * it returns -enospc
 */
int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2492
			     struct btrfs_root *root,
2493 2494 2495 2496
			     struct btrfs_block_group_cache *block_group,
			     struct btrfs_free_cluster *cluster,
			     u64 offset, u64 bytes, u64 empty_size)
{
2497
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2498
	struct btrfs_free_space *entry, *tmp;
2499
	LIST_HEAD(bitmaps);
2500
	u64 min_bytes;
2501
	u64 cont1_bytes;
2502 2503
	int ret;

2504 2505 2506 2507 2508 2509
	/*
	 * Choose the minimum extent size we'll require for this
	 * cluster.  For SSD_SPREAD, don't allow any fragmentation.
	 * For metadata, allow allocates with smaller extents.  For
	 * data, keep it dense.
	 */
2510
	if (btrfs_test_opt(root, SSD_SPREAD)) {
2511
		cont1_bytes = min_bytes = bytes + empty_size;
2512
	} else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2513 2514 2515 2516 2517 2518
		cont1_bytes = bytes;
		min_bytes = block_group->sectorsize;
	} else {
		cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
		min_bytes = block_group->sectorsize;
	}
2519

2520
	spin_lock(&ctl->tree_lock);
2521 2522 2523 2524 2525

	/*
	 * If we know we don't have enough space to make a cluster don't even
	 * bother doing all the work to try and find one.
	 */
2526
	if (ctl->free_space < bytes) {
2527
		spin_unlock(&ctl->tree_lock);
2528 2529 2530
		return -ENOSPC;
	}

2531 2532 2533 2534 2535 2536 2537 2538
	spin_lock(&cluster->lock);

	/* someone already found a cluster, hooray */
	if (cluster->block_group) {
		ret = 0;
		goto out;
	}

2539
	ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2540 2541
				      bytes + empty_size,
				      cont1_bytes, min_bytes);
2542
	if (ret)
2543
		ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2544 2545
					   offset, bytes + empty_size,
					   cont1_bytes, min_bytes);
2546 2547 2548 2549

	/* Clear our temporary list */
	list_for_each_entry_safe(entry, tmp, &bitmaps, list)
		list_del_init(&entry->list);
2550

2551 2552 2553 2554 2555
	if (!ret) {
		atomic_inc(&block_group->count);
		list_add_tail(&cluster->block_group_list,
			      &block_group->cluster_list);
		cluster->block_group = block_group;
2556 2557 2558
	}
out:
	spin_unlock(&cluster->lock);
2559
	spin_unlock(&ctl->tree_lock);
2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570

	return ret;
}

/*
 * simple code to zero out a cluster
 */
void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
{
	spin_lock_init(&cluster->lock);
	spin_lock_init(&cluster->refill_lock);
2571
	cluster->root = RB_ROOT;
2572 2573 2574 2575 2576
	cluster->max_size = 0;
	INIT_LIST_HEAD(&cluster->block_group_list);
	cluster->block_group = NULL;
}

2577 2578 2579
int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
			   u64 *trimmed, u64 start, u64 end, u64 minlen)
{
2580
	struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2581 2582 2583 2584 2585 2586 2587 2588 2589
	struct btrfs_free_space *entry = NULL;
	struct btrfs_fs_info *fs_info = block_group->fs_info;
	u64 bytes = 0;
	u64 actually_trimmed;
	int ret = 0;

	*trimmed = 0;

	while (start < end) {
2590
		spin_lock(&ctl->tree_lock);
2591

2592 2593
		if (ctl->free_space < minlen) {
			spin_unlock(&ctl->tree_lock);
2594 2595 2596
			break;
		}

2597
		entry = tree_search_offset(ctl, start, 0, 1);
2598
		if (!entry)
2599 2600
			entry = tree_search_offset(ctl,
						   offset_to_bitmap(ctl, start),
2601 2602 2603
						   1, 1);

		if (!entry || entry->offset >= end) {
2604
			spin_unlock(&ctl->tree_lock);
2605 2606 2607 2608
			break;
		}

		if (entry->bitmap) {
2609
			ret = search_bitmap(ctl, entry, &start, &bytes);
2610 2611
			if (!ret) {
				if (start >= end) {
2612
					spin_unlock(&ctl->tree_lock);
2613 2614 2615
					break;
				}
				bytes = min(bytes, end - start);
2616
				bitmap_clear_bits(ctl, entry, start, bytes);
2617
				if (entry->bytes == 0)
2618
					free_bitmap(ctl, entry);
2619 2620 2621
			} else {
				start = entry->offset + BITS_PER_BITMAP *
					block_group->sectorsize;
2622
				spin_unlock(&ctl->tree_lock);
2623 2624 2625 2626 2627 2628
				ret = 0;
				continue;
			}
		} else {
			start = entry->offset;
			bytes = min(entry->bytes, end - start);
2629
			unlink_free_space(ctl, entry);
2630
			kmem_cache_free(btrfs_free_space_cachep, entry);
2631 2632
		}

2633
		spin_unlock(&ctl->tree_lock);
2634 2635

		if (bytes >= minlen) {
2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648
			struct btrfs_space_info *space_info;
			int update = 0;

			space_info = block_group->space_info;
			spin_lock(&space_info->lock);
			spin_lock(&block_group->lock);
			if (!block_group->ro) {
				block_group->reserved += bytes;
				space_info->bytes_reserved += bytes;
				update = 1;
			}
			spin_unlock(&block_group->lock);
			spin_unlock(&space_info->lock);
2649 2650 2651 2652 2653 2654

			ret = btrfs_error_discard_extent(fs_info->extent_root,
							 start,
							 bytes,
							 &actually_trimmed);

2655
			btrfs_add_free_space(block_group, start, bytes);
2656 2657 2658 2659 2660 2661 2662 2663 2664 2665
			if (update) {
				spin_lock(&space_info->lock);
				spin_lock(&block_group->lock);
				if (block_group->ro)
					space_info->bytes_readonly += bytes;
				block_group->reserved -= bytes;
				space_info->bytes_reserved -= bytes;
				spin_unlock(&space_info->lock);
				spin_unlock(&block_group->lock);
			}
2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683

			if (ret)
				break;
			*trimmed += actually_trimmed;
		}
		start += bytes;
		bytes = 0;

		if (fatal_signal_pending(current)) {
			ret = -ERESTARTSYS;
			break;
		}

		cond_resched();
	}

	return ret;
}
2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733

/*
 * Find the left-most item in the cache tree, and then return the
 * smallest inode number in the item.
 *
 * Note: the returned inode number may not be the smallest one in
 * the tree, if the left-most item is a bitmap.
 */
u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
{
	struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
	struct btrfs_free_space *entry = NULL;
	u64 ino = 0;

	spin_lock(&ctl->tree_lock);

	if (RB_EMPTY_ROOT(&ctl->free_space_offset))
		goto out;

	entry = rb_entry(rb_first(&ctl->free_space_offset),
			 struct btrfs_free_space, offset_index);

	if (!entry->bitmap) {
		ino = entry->offset;

		unlink_free_space(ctl, entry);
		entry->offset++;
		entry->bytes--;
		if (!entry->bytes)
			kmem_cache_free(btrfs_free_space_cachep, entry);
		else
			link_free_space(ctl, entry);
	} else {
		u64 offset = 0;
		u64 count = 1;
		int ret;

		ret = search_bitmap(ctl, entry, &offset, &count);
		BUG_ON(ret);

		ino = offset;
		bitmap_clear_bits(ctl, entry, offset, 1);
		if (entry->bytes == 0)
			free_bitmap(ctl, entry);
	}
out:
	spin_unlock(&ctl->tree_lock);

	return ino;
}
2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751

struct inode *lookup_free_ino_inode(struct btrfs_root *root,
				    struct btrfs_path *path)
{
	struct inode *inode = NULL;

	spin_lock(&root->cache_lock);
	if (root->cache_inode)
		inode = igrab(root->cache_inode);
	spin_unlock(&root->cache_lock);
	if (inode)
		return inode;

	inode = __lookup_free_space_inode(root, path, 0);
	if (IS_ERR(inode))
		return inode;

	spin_lock(&root->cache_lock);
2752
	if (!btrfs_fs_closing(root->fs_info))
2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774
		root->cache_inode = igrab(inode);
	spin_unlock(&root->cache_lock);

	return inode;
}

int create_free_ino_inode(struct btrfs_root *root,
			  struct btrfs_trans_handle *trans,
			  struct btrfs_path *path)
{
	return __create_free_space_inode(root, trans, path,
					 BTRFS_FREE_INO_OBJECTID, 0);
}

int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
{
	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	struct btrfs_path *path;
	struct inode *inode;
	int ret = 0;
	u64 root_gen = btrfs_root_generation(&root->root_item);

C
Chris Mason 已提交
2775 2776 2777
	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
		return 0;

2778 2779 2780 2781
	/*
	 * If we're unmounting then just return, since this does a search on the
	 * normal root and not the commit root and we could deadlock.
	 */
2782
	if (btrfs_fs_closing(fs_info))
2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815
		return 0;

	path = btrfs_alloc_path();
	if (!path)
		return 0;

	inode = lookup_free_ino_inode(root, path);
	if (IS_ERR(inode))
		goto out;

	if (root_gen != BTRFS_I(inode)->generation)
		goto out_put;

	ret = __load_free_space_cache(root, inode, ctl, path, 0);

	if (ret < 0)
		printk(KERN_ERR "btrfs: failed to load free ino cache for "
		       "root %llu\n", root->root_key.objectid);
out_put:
	iput(inode);
out:
	btrfs_free_path(path);
	return ret;
}

int btrfs_write_out_ino_cache(struct btrfs_root *root,
			      struct btrfs_trans_handle *trans,
			      struct btrfs_path *path)
{
	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
	struct inode *inode;
	int ret;

C
Chris Mason 已提交
2816 2817 2818
	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
		return 0;

2819 2820 2821 2822 2823
	inode = lookup_free_ino_inode(root, path);
	if (IS_ERR(inode))
		return 0;

	ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2824 2825 2826
	if (ret) {
		btrfs_delalloc_release_metadata(inode, inode->i_size);
#ifdef DEBUG
2827 2828
		printk(KERN_ERR "btrfs: failed to write free ino cache "
		       "for root %llu\n", root->root_key.objectid);
2829 2830
#endif
	}
2831 2832 2833 2834

	iput(inode);
	return ret;
}